Joint source-channel decoding approaches applied in adaptive multirate (AMR) speech transmission

TLDR: Investigations are reported on joint source-channel (JSC) decoding applied in the adaptive multirate (AMR) speech codec for typical real speech, and it is shown that for the AMR codec, although highly optimized in exploiting signal redundancy, some redundancy still remains in the encoded speech bits.

Abstract: Investigations are reported on joint source-channel (JSC) decoding applied in the adaptive multirate (AMR) speech codec for typical real speech. The studied approaches include the "HUK" method, two-step decoding, repeated decoding, and error concealment based on the so-called soft bit decoding. A method based on a Kalman filter for a priori information estimation is described, where both inter- and intra-frame correlation are taken into account. It is shown that for the AMR codec, although highly optimized in exploiting signal redundancy, some redundancy still remains in the encoded speech bits. By using an optimized JSC decoding algorithm like the repeated decoding, a gain of up to 0.2 dB in channel signal-to-noise ratio is achievable, compared to decoding without exploiting the residual redundancy. Under bad channel conditions, the error concealment algorithm can deliver an even better subjective speech quality.